Investigation Quickstart
You do not need to understand the whole project to begin investigating it.
The Acts of Erudition and the research materials surrounding it can get dense quickly. Terms are defined precisely because the experiment depends on precision—but you should not need to learn an entire vocabulary before you can determine what is being tested or decide how you might contribute useful evidence.
This page is the short version.
The experiment asks a relatively simple question:
Can a bounded written artifact preserve enough of a person’s reasoning architecture for an AI model to reconstruct how that person reasons about genuinely new problems?
The current answer is unresolved.
The materials on this site are designed to make that question testable rather than merely philosophical.
Public access and experimental access are different.
Anyone can use the public site to understand the hypothesis, published methodology, limitations, and developing research record. Formal confirmatory results are not yet published. The complete Scientific Owner Library requires owner verification with the physical book. Experimental roles receive only their assigned materials from the Test Administrator. Question Authors and blind raters do not need to read the Book; doing so can make them ineligible for their role. Direct research participation by Austin is separate and must be arranged independently.
What You Need
What you need depends on your activity and assigned role; the materials below are not prerequisites for every participant.
The Book
The Acts of Erudition is the human-authored artifact being tested. It contains the distinctions, priorities, assumptions, tensions, standards of justification, and methods of revision that the reconstruction process attempts to recover. The book is not an answer key. It is the experimental object.
An AI Model
The experiment is not dependent on one specific model. A compatible language model receives the bounded artifact and the Ask the Book instructions, then attempts to reconstruct the reasoning represented by the artifact when answering questions that were not already resolved within it. Using different models is useful because it helps distinguish what may be coming from the artifact from what may be peculiar to one model.
The Research Package
The Scientific Owner Library contains the protected materials used to conduct, score, document, or independently examine the experiment. Access to the complete owner library requires owner verification. Confirmatory Question Authors and blind Raters receive only their assigned materials from the Test Administrator; they do not need to purchase the Book, verify ownership, or enter the complete Scientific Owner Library to participate. Public visitors can still examine the hypothesis, published methodology, limitations, and public research record without access to the protected package.
These protected materials include the LLM edition, prompt, protocol, scoring materials, rater materials, data workbook, analysis specification, preregistration materials, version manifest, and other supporting files.
You do not need every file for every role.
Choose Your Role
There are three primary experimental roles. These roles require access to the materials assigned to them by the applicable protocol or administrator. Public visitors do not need access to these protected materials to learn how the roles work or examine the developing research record; formal confirmatory results are not yet published.
1. Rater
What you do: Apply the frozen eight-dimensional Structural Continuity Index (SCI) to the original question and anonymized response pairs supplied by the Test Administrator. The protocol requires raters not to see condition labels, model identity, which response is Austin’s, the study hypothesis, or prior scores. Raters must not use the complete owner library or infer identities while rating.
You do not need to design questions, administer the experiment, or decide whether the hypothesis is true.
Your job is narrower: apply the scoring rules consistently. This public site describes the hypothesis, so reading it may conflict with the protocol’s hypothesis-blinding requirement. The Test Administrator must screen and document prior project knowledge and recruit eligible independent raters; public recruitment alone does not guarantee blinding.
Time commitment: A single rating task should be relatively short. A larger rating set will take longer depending on the number and complexity of responses.
Why your role matters
A reconstruction claim means very little if the person making the claim also decides whether every reconstruction succeeded.
Independent raters create distance between the hypothesis and its evaluation.
A disagreement is useful data too.
2. Question Author
What you do: Create genuinely novel questions. Under frozen v1.0, Question Authors must not have read the manuscript or been briefed on Austin’s likely answers. They may receive the authoring packet and domain matrix, but not the manuscript, Austin’s answers, model outputs, or expected conclusions. The Test Administrator checks contamination and freezes question identities before Austin or the model sees them. Creating a question does not authorize presenting it to Austin; new human-reference exposure must be arranged through the Scientific / Research Participation pathway.
Good questions should require reasoning rather than simple retrieval from the book.
The objective is not to write questions the system is likely to answer well.
It is to create questions capable of distinguishing genuine structural reconstruction from quotation, imitation, generic good advice, or lucky agreement.
Time commitment: Preparing a rigorous question bank takes time. The Test Administrator checks prior exposure and overlap with the artifact, pilot bank, website, interviews, and earlier tests without giving Question Authors access to excluded material.
Why your role matters
The experiment becomes weak if the questions accidentally contain answers the book has already supplied.
Novel questions force the reconstruction to move beyond retrieval.
They are where the central claim is actually tested.
3. Test Administrator
What you do: Run the procedure while preserving the boundaries of the test.
This includes supplying role-specific materials, maintaining separation and required blinding, freezing questions and conditions, collecting independent T1/T2 human-reference responses, preserving anonymized response mappings, recording exclusions, and following the frozen primary-analysis and secondary-testing sequence.
Time commitment: Administration requires the largest commitment. A limited trial may take a few hours; a larger confirmatory run can require substantially more time across preparation, administration, and documentation.
Why your role matters
Even a good hypothesis can produce meaningless evidence if the procedure changes while it is being tested.
The administrator protects the experiment from becoming an informal demonstration.
The Process, in Order
For a formal test, the basic sequence is:
1. Freeze the experimental materials. Establish exactly which version of the artifact, prompt, protocol, scoring system, and question bank will be used.
2. Create or select novel questions. Questions should test reasoning the artifact has not already explicitly resolved.
3. Run the reconstruction. Provide the bounded materials to the selected model under the specified conditions and collect its responses.
4. Obtain the human reference. Where the study design requires a new human-reference response, Austin answers the same novel questions independently under the conditions specified by the protocol. Owner access does not include access to Austin as a research participant. Studies requiring new participation by Austin must arrange that participation separately through the Scientific / Research Participation pathway before exposure occurs.
5. Prepare the comparison. Responses and identifying information are organized according to the blinding and administration procedures.
6. Rate the responses. Independent raters apply the scoring manual rather than inventing their own definition of success.
7. Record the results. Scores, disagreements, exclusions, missing data, procedural deviations, and other required information are preserved.
8. Analyze the completed trial. Only after the planned testing is complete should the aggregate evidence be interpreted.
A failed reconstruction is not something to repair during the experiment.
It is a result.
What Happens to Your Name and Data?
Participation should not require unnecessary public disclosure of your identity.
Research records should distinguish between information necessary to evaluate the experiment and information that merely identifies the person who contributed it.
Where a name is not necessary to the evidentiary record, the working data can identify participants by role or participant code instead.
If public attribution is proposed—for example, identifying a question author, rater, administrator, collaborator, institution, or contributor—that should be handled separately from the underlying research data and with the participant’s knowledge.
I do not want disagreement hidden.
If you rate a reconstruction poorly, identify a methodological problem, produce a failed replication, or challenge the hypothesis, that is relevant evidence rather than disloyal participation.
The point is to find out what survives scrutiny.
Conflict-of-Interest Disclosure
This project has an unavoidable conflict of interest that should be understood before evaluating its results.
I am the author of The Acts of Erudition, the person whose reasoning the current artifact attempts to reconstruct, and the creator and primary organizer of the Ask the Book experiment. I developed the underlying framework, assembled the research materials, and have a personal, intellectual, reputational, and potentially financial interest in the project receiving attention and producing meaningful results.
Positive findings could benefit me through book sales, research opportunities, speaking or media opportunities, future products, licensing or commercial development, and recognition of the methodology. I therefore should not be treated as a disinterested evaluator of my own hypothesis.
The project is structured to reduce—but cannot eliminate—this conflict. The experimental artifact and specified materials are versioned and frozen during testing; scoring procedures are documented in advance; independent question authors, administrators, and raters can participate; methodological criticism and negative results are permitted and should be preserved; and the project identifies the central hypothesis as unresolved rather than established.
My participation is also necessary for some current tests because the experiment compares reconstructions against my independently produced responses to genuinely novel questions. That makes me part of the experimental object as well as the person who initiated the research. This limitation should be considered when interpreting any result.
Independent replication, administration, rating, analysis, and eventual testing with additional people are therefore especially important. Evidence produced solely under my direction should not be treated as equivalent to evidence reproduced independently.
Nothing on this site should be interpreted as independent validation merely because I designed procedures intended to make independent validation possible.
Reconciliation and the Author's Dual Role
There is an additional complication: the author’s involvement is not solely a practical limitation of this first experiment. It is also part of the methodology being investigated.
One working hypothesis behind this project is that a mature deliberative artifact cannot be produced simply by observing a person from the outside and inferring patterns from their behavior or previous answers. The artifact and its maker may instead need to develop recursively: the person externalizes a reasoning structure, encounters that structure as an artifact, identifies distortions, omissions, contradictions, or aspirational representations within it, and revises until the representation is one they recognize as accurately expressing the architecture through which they actually reason.
I refer to this process as reconciliation.
Under this hypothesis, reconciliation between a deliberative artifact and its maker is not an undesirable contamination that could simply be removed from artifact construction. It is part of what may be required to produce the object being tested in the first place.
That does not remove the resulting conflict of interest.
The same person whose participation may be necessary to establish a reconciled artifact can also consciously or unconsciously shape that artifact toward a preferred self-description, influence experimental decisions, or interpret ambiguous results favorably. The methodology therefore creates a deliberate separation between constructing and reconciling the artifact and independently evaluating what that artifact can actually do.
The current experiment should consequently be understood as testing a reconciled, person-specific artifact—not as demonstrating that reconciliation itself has been validated as the necessary or optimal method for producing such artifacts.
Whether reconciliation is genuinely necessary, how much is necessary, whether it can be operationalized reliably, and whether the process generalizes beyond this first author-artifact pair remain separate empirical questions.
What I Will Respond To
I may respond to reasonable clarification questions necessary to understand the artifact, materials, procedure, authorship, or intended meaning of something I wrote. Clarification is not the same as participation by the human reference.
Questions that expose me to novel experimental material, require me to generate new research responses, involve substantial individualized work, or otherwise make my participation part of a study must use the appropriate Participation pathway. Scientifically relevant exposure may also need to become part of the experimental record.
I will clarify:
what a passage or project-specific term means;
where a relevant definition or document can be found;
how the materials are intended to be used;
factual questions about the artifact and its development;
ambiguities that prevent someone from conducting the procedure;
suspected errors, inconsistencies, or methodological problems;
requests necessary to coordinate legitimate participation or investigation.
I will also answer criticism.
Agreement with the hypothesis is not a condition of owner access, reasonable clarification, or methodological criticism.
What I Will Not Do
I will not privately coach a model, rater, question author, or administrator toward a preferred result.
I will not tell raters what score I want.
I will not help redesign a confirmatory question because the reconstruction performed poorly on it.
I will not reinterpret every failure until it becomes a success.
I will not change the frozen experimental artifact during a testing phase to improve its performance.
And I cannot provide unlimited informal consultation.
This project is currently being conducted alongside my ordinary work and family responsibilities. Until the research itself becomes something closer to my occupation, substantial individualized participation may require formal arrangements.
There is no expected or mandatory amount of involvement from anyone investigating the project. Depending on access and role, a single carefully constructed question, one independent rating session, a methodological criticism, or a complete replication can each be useful.
You Can Ignore the Jargon at First
You may encounter terminology elsewhere on the site that is unfamiliar on first reading.
That terminology matters when its precision becomes relevant.
It does not need to be your entrance requirement.
If you encounter something like disciplined nonfinality, a particular QM/QF configuration, or another project-specific term, you do not need to master the entire conceptual system before beginning your investigation.
Start with the operational question:
What am I being asked to do, and what evidence would this step produce?
Then consult the formal definitions when they become necessary.
The detailed materials exist so that precision is available—not so that vocabulary becomes a barrier to entry.
The Standard Is Not Agreement
You do not need to believe the hypothesis.
In fact, skepticism is useful.
A strong investigation should be capable of producing evidence that the artifact works, works only under limited conditions, works for some dimensions but not others, fails to generalize, or does not work at all.
The experiment becomes interesting only if those outcomes can be distinguished.
The purpose of the research package is therefore not to teach someone how to reproduce my preferred conclusion.
It is to make the claim inspectable.
Start Small
If you are exploring publicly, start with the hypothesis, methodology, falsification criteria, and published methodology. Inspect the public account of the protocol for weaknesses. Compare the claim against adjacent research. Look for assumptions, uncontrolled variables, alternative explanations, or methodological problems. You do not need owner access to understand or criticize the experiment.
If you have verified owner access, you can go further by inspecting the protected protocol and scoring materials, running the bounded artifact through another compatible model, conducting exploratory work, administering eligible studies, or attempting an independent replication. Confirmatory Question Authors and blind Raters are different: eligibility for those roles depends on preserving the exposure and blinding requirements of the frozen protocol, not on owner verification. The Test Administrator must recruit and supply eligible participants with only the materials appropriate to their role.
If your investigation requires Austin as the human reference, owner access is not sufficient. Use the Scientific / Research Participation pathway before exposing him to novel experimental material or requesting new human-reference responses.
You do not have to understand everything before beginning to investigate the experiment. Sometimes the fastest way to understand a claim is to determine what would break it.
That is what these materials are designed to make possible.